[1/3] Optimize Slime Update Weights: Remove QWen3MOE Load Weight Overhead

[hebiao064]

Motivation

• 1/3: [1/3] Optimize Slime Update Weights: Remove QWen3MOE Load Weight Overhead #8751 （23s => 22s)
• 2/3: [2/3] Optimize Slime Update Weights: Avoid GPU-to-CPU Device Sync when update expert weights #8753 (22s => 12s)
• 3/3: [3/3] Optimize Slime Update Weights: Load Weight in non-blocking mode #8754 (12s => 10s)

This PR maps to the ASYNC_EP8_optimization_1 which save 2s

Modifications

Accuracy Test

GSM Before this PR:

Accuracy: 0.935
Invalid: 0.000
Latency: 13.829 s
Output throughput: 1653.462 token/s

GSM After this PR:

Accuracy: 0.935
Invalid: 0.000
Latency: 13.648 s
Output throughput: 1692.839 token/s

Benchmark & Profiling

Checklist

• Format your code according to the Code Formatting with Pre-Commit.
• Add unit tests as outlined in the Running Unit Tests.
• Update documentation / docstrings / example tutorials as needed, according to Writing Documentation.
• Provide throughput / latency benchmark results and accuracy evaluation results as needed, according to Benchmark and Profiling and Accuracy Results.
• For reviewers: If you haven't made any contributions to this PR and are only assisting with merging the main branch, please remove yourself as a co-author when merging the PR.
• Please feel free to join our Slack channel at https://slack.sglang.ai to discuss your PR.

[gemini-code-assist]

Summary of Changes

Hello @hebiao064, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request is the first in a series of three, focusing on optimizing the weight loading process for QWen3MOE models. My primary goal is to remove the overhead associated with loading weights, specifically targeting the load_weights method to improve its efficiency. The changes aim to reduce the time taken for model initialization, which is expected to save approximately 2 seconds, as indicated by the ASYNC_EP8_optimization_1 mapping.

Highlights

• Parameter Dictionary Caching: I've implemented a caching mechanism for self.named_parameters() by storing it in _cached_params_dict. This significantly reduces the overhead of repeatedly traversing the model's parameter tree during weight loading, as named_parameters() can be an expensive operation.
• Optimized Expert Weight Handling: I've added an is_expert_weight flag and corresponding logic to allow for early skipping of expert weights that are not relevant to the current processing rank. This prevents unnecessary processing and lookups for weights that won't be loaded by the current device, improving efficiency.
• Lazy Initialization of Routed Expert Weights: I've changed the initialization of self.routed_experts_weights_of_layer to be lazy. This means the computation of these weights is deferred until they are actually accessed, avoiding a potentially costly operation during the initial load_weights call if not immediately required.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in issue comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments or fill out our survey to provide feedback.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces several optimizations to the load_weights method in Qwen3MoeForCausalLM. These include caching the parameters dictionary and lazily initializing the expert weights cache to avoid expensive recomputations on repeated calls. It also adds logic to more efficiently skip loading expert weights that do not belong to the current rank. The changes are sound and contribute to the goal of reducing weight loading overhead. I have one suggestion to make the weight loading logic more robust against potential bugs.

[zhaochenyang20]

🐂🍺

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!